tipsv2-b14-torch#

TIPSv2 base vision-language encoder for zero-shot image classification and spatially-aware embeddings.

Details

  • Model name: tipsv2-b14-torch

  • Model source: https://huggingface.co/google/tipsv2-b14

  • Model author: Google DeepMind

  • Model license: Apache 2.0

  • Model size: 748.22 MB

  • Exposes embeddings? yes

  • Tags: classification, logits, embeddings, torch, transformers, zero-shot, official

Requirements

  • Packages: torch, torchvision, transformers>=5.14, sentencepiece

  • CPU support

    • yes

  • GPU support

    • yes

Example usage

 1import fiftyone as fo
 2import fiftyone.zoo as foz
 3
 4dataset = foz.load_zoo_dataset(
 5    "coco-2017",
 6    split="validation",
 7    dataset_name=fo.get_default_dataset_name(),
 8    max_samples=50,
 9    shuffle=True,
10)
11
12
13classes = ["person", "dog", "cat", "bird", "car", "tree", "chair"]
14
15model = foz.load_zoo_model(
16    "tipsv2-b14-torch",
17    classes=classes,
18)
19
20dataset.apply_model(model, label_field="predictions")
21
22session = fo.launch_app(dataset)